[Spambayes] Spam overload

[bill parducci]

i have had good luck using a shared hammiedb with a dozen users (a far 
cry from 7000!) only the administrator retrains in this situation. 
accuracy is not what it would be with a user centric model, but spam by 
its very nature is fairly common so intuition (and my limited user 
community feedback) suggests that false positives are where you see the 
most significant reduction in accuracy, and even then it isn't 
suprisingly good. if you are being 'overrun', then 90+% accuracy is 
still better than nothing. :)

while my users are imap based (procmail drops spam into each users SPAM 
folder on the server), it seems that a proxy solution would work if you 
did subject rewrites and let the downstream client filter on that.

personally, i would think thrice before implementing the outlook client 
on 7000 machines. on the other hand, you are not going to be able to 
have user trained hammiedbs with exchange otherwise (a proxy model 
wouldn't have access to the mboxes to train upon, so you would have to 
build a single message retrain model; easily doable on the server 
side--as long as you have the horsepower--but would require some 
significant user training)

have you considered imap? :-P

b